<?xml version="1.0" standalone="yes"?>
<Paper uid="J99-4002">
  <Title>Lexical Rules in Constraint-based Grammars</Title>
  <Section position="8" start_page="521" end_page="522" type="concl">
    <SectionTitle>
8. Conclusions
</SectionTitle>
    <Paragraph position="0"> Both Goldberg (1995) and Jackendoff (1997a, 1997b) contrast the lexical rule approach to bounded dependencies with one that treats each construction independently and characterizes relations between constructions, somewhat vaguely, in terms of &amp;quot;inheritance.&amp;quot; Jackendoff (1997b, 556f.) makes the point that lexical rules in lexicalist frameworks are expressive enough to describe bounded constructions and any idiosyncratic meanings they convey, so they can be used to capture relations among such constructions. He argues against this approach though, because he suggests that the elements of the constructions related by such rules are often not lexical. Copestake and Briscoe (1995) make the same point with respect to examples of systematic metonymy, where such semiproductive &amp;quot;lexical&amp;quot; rules apply to noun phrase constructions. Jackendoff also argues, however, that constructions need to be treated as a kind of phrasal lexical item whose (idiosyncratic) meaning is learnt like that of a lexical item (Jackendoff 1997b, 554). For us, the defining characteristic of a lexical rule is that it requires some listing of properties to accurately express its behavior, whether this be because it is lexically governed, has exceptions, is underspecified in its effects, or whatever. Thus idioms must be lexically specified, though they are best treated as particularly idiosyncratic phrases/constructions, rather than lexical items.</Paragraph>
    <Paragraph position="1"> For Jackendoff (1997a, 115f.), the crucial distinction is not lexical/nonlexical but productive/semiproductive rule. Jackendoff's definition of a productive rule encompasses rules such as Plural Noun Formation or Third Singular Verb Formation (see Section 2 above), despite the existence of irregular derived forms, because he argues that such a rule's output need not be listed. It is semiproductive only to the extent that certain aspects of its output can be overridden or blocked by lexical specification of exceptions. He reserves the term semiproductive for rules, such as Denominal Verb Formation (shelf ~-+ shelve), where the exact output of the rule is underspecified and the existence of the derived words is not guaranteed. Thus, the precise meaning of the denominal verb is partly systematic ('to put x in/on y') and partly idiosyncratic and unpredictable (e.g., to saddle (a horse) means 'to put a saddle on a horse's back,' while to shelve a book means 'to put a book on a shelf'), and the phonological form is not always identical with or entirely predictable from the nominal form (e.g., shelve). Furthermore, there are many nouns that do not have corresponding denominal forms (mustard vs.</Paragraph>
    <Paragraph position="2"> butter, teapot vs. knife, etc.). Jackendoff argues that the nature of the exceptions to the latter type of rule requires (full) listing in the lexicon of the derived forms, while the former does not.</Paragraph>
    <Paragraph position="3"> In our approach, lexical rules are those that require some element of lexical/listed specification, whether it be the listing of irregular forms that override aspects of the rule output or of idiosyncratic aspects of the resulting meaning, or the unattested status of the derived entry. The approach to lexical rules we have advocated, integrating a restrictive default-based formalization with partial precompilation and a probabilistic account of item-familiarity and semiproductivity, is capable of expressing inflectional, derivational, and conversion rules whose domain is (within) that of a bounded dependency construction (i.e., includes &amp;quot;alternation&amp;quot; rules relating bounded constructions). This approach reintegrates construction-based generalizations with more traditional  Briscoe and Copestake Lexical Rules lexical rules, provides a very general means for encoding semiproductivity, and makes a principled distinction between lexical and unary syntactic rules that should allow the generative power of the overall grammar to be restricted. To summarize: lexical rules cannot perform arbitrary operations on unbounded lists; they are unidirectional, but have limited reversibility properties appropriate to account for backformation; they involve no extension to the underlying logic of the TDFS framework; they require the statement of what changes, not what stays the same; they are subject to type constraints and can exploit the default inheritance hierarchy to capture generalizations; they can be semiproductive and are predicted to be sensitive to blocking, exceptions, conventionalization, and so forth; and they allow a linguistically elegant, and accurate, account of the dative construction/alternation, subsuming the insights emerging from recent detailed analyses of this specific lexical rule.</Paragraph>
    <Paragraph position="4"> Furthermore, we outlined ways in which this approach can be extended straight-forwardly to deal with rules apparently involving more complex SUBCAT list manipulations, and with recursive processes of derivation and conversion and their associated semantics, without sacrificing these desirable properties. It follows from our approach that some putative lexical rules should be treated as unary syntactic rules. A potential advantage of this division of labor is that it may even be possible to develop a separate treatment of unary syntactic rules that does not utilize category-valued variables over list-valued features. In any case, if these rules only apply to the output of the lexicon, this will avoid the increase in generative capacity identified by Carpenter (1991), resulting from the interaction of recursion, arbitrary list operations and unbounded lists, by keeping list-valued features bounded during lexical rule application, and only allowing unbounded additions to, or limited modification of, such features during syntactic processing.</Paragraph>
  </Section>
class="xml-element"></Paper>